摘要
仿射传播(Affinity propagation,AP)聚类算法是将所有待聚类对象作为潜在的聚类中心,通过对象之间传递的可靠性和有效性信息找到合适的聚类中心,从而计算出相应的聚类结果,但不适用子空间聚类。将粒度计算引入到仿射传播聚类算法中,提出属性与样本同步粒化的AP熵加权软子空间聚类算法(Entropy weighting AP algorithm for subspace clustering based on asynchronous granulation of attributes and samples,EWAP)。EWAP首先去除冗余属性,然后在每次聚类的迭代过程中修改属性的权重值。在满足一定条件迭代终止时,就会得到构成各兴趣度子空间的属性权重值,从而得到属性集的粒化结果以及相应的子空间聚类结果。理论与实验证明EWAP算法既保留了AP算法的优点,又克服了该聚类算法不能进行子空间聚类的不足。
Affinity propagation （AP） clustering algorithm considers all clustering objects as potential clustering centers, and messages of responsibility and availability are exchanged between objects until a high- quality set of clustering centers and corresponding clusters gradually emerge. But it is not appropriate for subspace clustering. To solve this problem, an entropy weighting AP algorithm for subspace clustering based on asynchronous granulation of attributes and samples （EWAP） is put forward through introducing the idea of granular computing into the affinity propagation clustering method. It removes the redundant attributes first, and then a step of modifying attribute weights is added to the clustering procedure for obtaining the exact weights value. At the end of iteration, the attribute weights of each subspace, an accu- rate result of attributes granularity and the corresponding clusters will be produced. The theory and practice prove that EWAP preserves the advantages of AP clustering and overcomes its shortage of unsatisfying subspace clustering.
出处
《数据采集与处理》
CSCD
北大核心
2016年第4期767-774,共8页
Journal of Data Acquisition and Processing
基金
国家自然科学基金(61379101)资助项目
江苏省自然科学基金(BK20130209)资助项目
江苏省高校自然科学基金(14KJB520039)资助项目
江苏省高校优秀中青年教师和校长境外研修计划资助项目
关键词
聚类
熵
加权
子空间
AP聚类
clustering
entropy
weighting
subspace
AP clustering
